A Lightweight YOLO-Based Algorithm for Digital Target Recognition Integrating Positional Enhancement and Morphological Processing

Target detection is vital for modern military applications, yet deploying deep learning models on resource-limited edge devices remains challenging. Existing lightweight models often exhibit poor boundary localization and low digit recognition accuracy, falling short of real-time and precision requirements. This paper introduces a lightweight YOLO-based algorithm enhanced with a novel positional mechanism and morphological processing. The key component, a position-enhanced feature pyramid network (Enhanced-FPN), fuses shallow high-resolution and deep semantic features to improve localization accuracy. A ShuffleNetv2 backbone ensures low computational overhead, while a postdetection module applies morphological processing to robustly extract digit contours and orientation. Evaluated on a custom military data set, the model achieves 49.92% mean average precision at a 50% intersection over union threshold (mAP50) at 11.24 frames/second on an edge device—improving accuracy by 7.02 points and speed by 12.8% over the Yolo-FastestV2 baseline, with a comparable 0.11 GFLOP (Giga Floating Point Operations per second) cost. These results highlight the method’s effectiveness for real-time, high-precision target recognition in constrained environments.